BanglaFakeNews: A Curated Dataset for Bengali Fake News Detection

Description

BanglaFakeNews is a large-scale, curated dataset developed for fake news detection in the Bengali language. The dataset contains 50,800 labeled Bengali news articles collected from multiple publicly available online news sources and verified fact-checking platforms. Each news article is manually assigned one of two labels: Fake or Authentic, enabling binary classification for misinformation detection. The dataset was created to address the scarcity of high-quality benchmark datasets for Bengali fake news research. It covers a diverse range of topics, including politics, health, entertainment, sports, technology, crime, education, business, and social issues, making it suitable for developing robust and generalizable machine learning and deep learning models. The text data have undergone preprocessing to improve consistency and usability. The preprocessing pipeline includes duplicate removal, text normalization, tokenization, stop-word removal, and stemming while preserving the semantic content of the original news articles. The dataset is provided in CSV format and is intended to facilitate reproducible research in natural language processing (NLP), misinformation detection, text classification, and artificial intelligence applications for low-resource languages. Researchers can use this dataset to develop, evaluate, and benchmark traditional machine learning models, transformer-based architectures, and hybrid deep learning frameworks for Bengali fake news detection. It may also support related research in sentiment analysis, misinformation analysis, domain adaptation, explainable AI, and multilingual NLP. If this dataset contributes to your research, please cite the associated publication: A Hybrid Deep Learning Framework for Fake News Detection in Bengali News.

Steps to reproduce

1. Download the BanglaFakeNews.csv dataset from Mendeley Data. 2. Load the dataset using Python (e.g., Pandas) or any compatible data analysis software. 3. Perform standard text preprocessing, including text normalization, tokenization, stop-word removal, and stemming, if required. 4. Split the dataset into training, validation, and testing subsets (e.g., 80:10:10 or 70:15:15) or use cross-validation. 5. Convert the text into suitable feature representations such as TF-IDF embeddings, Word2Vec, FastText, or contextual embeddings from transformer models. 6. Train and evaluate machine learning or deep learning models for binary classification of Fake and Authentic news articles. 7. Report evaluation metrics such as Accuracy, Precision, Recall, F1-score, and AUC for comparison with future studies.

Institutions

• University of Barishal

  Barisal Division, Barishal

Categories

Licence